1. The supplementary eye field (SEF) has been viewed as a premotor cortical field for the selection and control of saccadic eye movements. Drawing on studies of the neighboring premotor cortex, we hypothesized that if the SEF participates in the selection of action based on arbitrary stimulus-response associations, then task-related activity in the SEF should change during the learning of such associations. 2. Rhesus monkeys were operantly conditioned to make a saccadic eye movement to one of four targets (7 deg up, down, left, and right from center) in response to a foveal instruction stimulus (IS). One and only one of those four possible responses was arbitrarily designated "correct" for each IS. The monkeys responded to familiar ISs, four stimuli that remained unchanged throughout training and recording, as well as to novel ISs, which the monkeys had not previously seen. The monkeys initially chose responses to novel stimuli by trial and error, with near chance levels of performance, but quickly learned to select the correct saccade. 3. We studied 186 SEF cells as monkeys learned new visuomotor associations. Neuronal activity was quantified in four task periods: during the presentation of the IS, during an instructed delay period (i.e., after the removal of the IS but before a trigger or "go" stimulus), just before the saccade, and after the saccade during fixation of the target location. The discharge rate in each task period was considered a separate case for analysis, compared with that in a reference period preceding the IS, and eliminated from further analysis if not significantly different. 4. We observed two main categories of activity change during learning, which we termed learning selective and learning dependent. Learning-selective cases showed a significant evolution in activity as the monkeys learned which saccade was instructed by a novel IS, but had no significant modulation during trials with familiar ISs. Many of these cells were virtually inactive on trials with familiar ISs. However, they initially showed dramatic modulation when tested with a novel IS. As the monkey chose the correct saccade (or target) with increasing reliability, the modulation often decremented until the cell was again relatively unmodulated, as observed during familiar-IS trials. These cells usually remained relatively inactive until the monkeys were challenged to start learning another new stimulus-response association. Learning-selective activity was observed in all task periods, and 33 (18%) of the 186 adequately tested SEF cells showed learning-selective activity in one or more task periods.(ABSTRACT TRUNCATED AT 400 WORDS)